File size: 877 Bytes
58b76f1
 
 
 
81fff32
d8c3a70
d87c146
58b76f1
 
 
81fff32
58b76f1
 
 
05bebc1
d8c3a70
58b76f1
d87c146
 
58b76f1
d8c3a70
d87c146
 
 
58b76f1
 
05bebc1
 
d87c146
 
 
05bebc1
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
# /// script
# requires-python = ">=3.10"
# dependencies = [
#     "numpy",
#     "torch==2.8.0",
#     "kernels",
#     "kernels-benchmark-tools",
# ]
#
# [tool.uv.sources]
# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
# ///
import torch
import sys
from kernels_benchmark_tools import KernelTypeEnum, run_benchmark
from kernels import get_kernel

# Load the sage attention kernel
hf_kernels_sage_attn = get_kernel("kernels-community/sage_attention")


def sage_attention(query, key, value):
    """SageAttention with INT8 Q/K quantization and FP16 P/V"""
    return hf_kernels_sage_attn.fwd(query, key, value, is_causal=False)[0]


run_benchmark(
    kernel_type=KernelTypeEnum.ATTENTION,
    impl_name="sage_int8_fp16",
    impl_tags={"family": "sageattention", "backend": "int8_fp16_cuda", "compile": "none"},
    impl_func=sage_attention,
)